Home

Row {Data-height = 350}

Participants

103

Countries

18

Songs

12

Row

How does the language of lyrics in music influence people’s perception of valence?

Column

Main results

Participants

Countries

Map with number of participants per country

Songs

Listen to the following songs:

A Hindi song with low valence:

Jagga Jiteya - Daler Mehndi, Dee MC, Shashwat Sachdev

A Dutch song with mid valence:

Treur niet (Ode aan het leven) - Diggy Dex, JW Roy

A Mandarin song with high valence:

Mojito - Jay Chou

Did you notice the differences in happiness between the songs? Did you enjoy the songs whether you can understand the language or not? we looked into whether the perceived happiness (or valence) level differs between a foreign language and one’s native language. This was done by asking questions like:

Did you enjoy listening to this song?

How happy or sad did this song make you feel?

Did you already know this song before listening?

Background

Column

Motivation

Our team consists of four members, namely Hannah, Mylène, Subu and Yili. While brainstorming for a topic for our project, it came to our attention that each of us speaks a different native language - Hannah is a native Dutch speaker, Mylène is a native English speaker, Subu is a native Hindi speaker, and Yili is a native Mandarin speaker. Inspired by this cultural diversity that is unique to our team, we became curious about how listeners who speak different native languages perceive music. Consequently, this research focused on the four populations of native speakers that each member of our team is a part of.

Based on our common interest in the area of music cognition within the domain of music information retrieval (MIR), we decided to focus on studying emotion perception in music. Combined with our interest in cross-cultural differences, we formulated the following research question: how differently do people perceive songs in a non-native language compared to songs in their native language?

Previous research have found evidence that basic emotions in Western music (Fritz et al., 2009) and pop music (Lee et al., 2021) are recognizable across cultures. However, to the best of our knowledge, no research has studied cross-cultural differences in perceived emotion in music comparing native Dutch, Hindi and Mandarin speakers with native English speakers. Our research contributes to literature by addressing this knowledge gap. The findings of our study could also be interesting for music marketers who would like to gain consumer insight into how listeners perceive songs in their native language versus songs that are in a non-native language.

This research project was conducted as part of the course Honoursmodule: The Data Science of Everyday Music Listening at the University of Amsterdam (coordinated by dhr. dr. John Ashley Burgoyne).

Column

Article Review: Cross-Cultural Mood Perception In Pop Songs And Its Alignment With Mood Detection Algorithms (Lee et al., 2021)

How differently do people perceive songs in a non-native language compared to songs in their native language? A recent study led by Lee et al. (2021) provides some insights into this question.

Putting their focus on mood perception in music specifically, Lee et al. (2021) examined three research questions in their study: first, to what extent people with different cultural backgrounds perceive mood in music differently; second, to what extent people within and between cultures agree on their perception of mood in music; and third, to what extent mood detection algorithms in Music Information Retrieval (MIR) accurately reflect human judgement, and whether a cultural bias can be found in the algorithm.

As an attempt to answer the research questions, the researchers analyzed three independent pools of participants from three countries (namely Brazil, South Korea and the US) that speak three different languages (namely Portuguese, Korean and English). The researchers’ explanation for selecting these populations is simple but rather convincing - there is evidence indicating that unique clusters in the West, Asia, and Latin America are formed due to shared music interests (notice how this is consistent with the top three globally most streamed artists on Spotify as well!).

The researchers’ choice of manually compiling a novel dataset of pop songs as experiment stimuli is refreshing, as most previous research (e.g. Eerola & Vuoskoski, 2011) utilized musical pieces that are known to evoke specific emotions. This makes the experiment setting feel more similar to real-world experiences, improving the external validity of the research findings.

A key finding suggests that emotions in pop songs may be universally recognizable to audiences regardless of cultural background, even when the song is unfamiliar. This statement was inferred based on the finding that listeners within and across cultures highly agreed on the presence of mood attributes such as danceable, energy, sad, cheerful and electronic in songs.

Another interesting result is the researcher’s conclusion that there is no cultural bias in the algorithms of Spotify API. Initially, the researchers hypothesized that the algorithm that was developed in the West would align better with participants in the US; however, statistical analyses revealed that there were no significant differences between the algorithm’s judgement and those of participants across cultures. Although this finding seems promising at first glance, the researchers’ arguments may not be robust enough to make such a conclusion, especially since the study did not include other cultures such as the African cultures and other Asian cultures.

Songs

Column

Song data

Title Artist Language Valence
Kuan Shu Y2j Mandarin 0.206
Respect Tino Martin Dutch 0.206
Don’t look back in anger Oasis English 0.206
Jagga Jiteya Daler Mehndi, Dee MC, Shashwat Sachdev Hindi 0.206
Ai De Jiu Shi Ni Leehom Wang Mandarin 0.571
Treur niet (Ode aan het leven) Diggy Dex, JW Roy Dutch 0.571
Maybe tomorrow Stereophonics English 0.571
Shubhaarambh Amit Trivedi, Shruti Pathak, Divya Kumar Hindi 0.571
Mojito Jay Chou Mandarin 0.800
Leef nu het kan Jan Smit Dutch 0.802
Walk of life Dire Straits English 0.802
Mummy Nu Pasand (From ‘Jai Mummy Di’) Sunanda Sharma, Tanishjk Bagchi Hindi 0.802

Survey

Column (data-width=300)

General questions of the Survey:
Question Scale
What is your age in years? e.g. 22 Years
What gender do you identify the most with? M/F/N-B/Other
What country are you from? List of Countries
What is the highest level of education that you have completed? Selected Choice
How fluent are you in Dutch/English/Hindi/Mandarin No proficiency- Native proficiency
How often do you listen to music in a non-native language? Never - Always
Questions after listening to each song:
Question Scale
How much did you enjoy listening to the song? Not at all - A great deal
How would you describe the song that you have just heard? Sad - Happy
How familiar are you with this song? Not familiar at all - Extremely familiar

Column (data-width=300)

Problems Encountered

  • While the spotify API provides its information in decimal points, due to a difficulty in direct comparison between Valence points by spotify and a scale rating by users, we decided to select tracks with the same valence points, and compare them to each other.

  • Moreover, due to happiness being subjective, based on past personal experiences, we needed to account for a lack of objectivity within our results.

  • Furthermore, we did not take into account the fact that a lot of Indian audiences consider their native language to be English instead of Hindi. This also resulted in a lower amount of participants with Hindi as a native language as compared to the English native speakers specifically.

Participants

Column

Language distribution

Age-gender distribution

Column

Participant statistics

In order to give some insight about the experiment, both graphs on the left provide some basic information about the 103 participants.

Language Distribution

The pie chart is a distribution of participants who have at least full professional proficiency in the languages tested in the experiment.

Note. Bilingual participants were counted multiple times. Participants in “Other” did not have at least full professional proficiency in any of the languages.

Age-gender distribution

The histogram shows the distribution of the participants in terms of age and gender. Most participants were female followed by males then non-binary and, lastly, there was one person categorised as ‘other’. The participants were on average around 27 years old and high school graduates.

Valence

Column

Valence as a measure for happiness

Valence is a rating that is given by sound engineers at Spotify which measures the positiveness of a song. High valence values indicate that a song is happier in nature, while now valence indicates sadder songs. To measure the “happiness” of a song, spotify data scientists have an agreed upon metric of what happy music sounds like. However according to Spotify a large amount of this measurement is also done by AI technology; therefore making it unclear exactly what the indicators of a happy song are. Spotify issues a measurement of valence in a decimal system, with values between 0 to 1 going up to three decimal points.

Psychologists on the other hand measure valence as the probability of a thing making people experience positive emotions.

General Observations

On the right, the total perceived valence can be seen. It seems that in general:

  • Participants mostly agreed on the perceived valence in songs regardless of language

  • Hindi songs were consistently rated high in perceived valence regardless of its valence rating on Spotify

  • People find it difficult to differentiate between sad and neutral songs, they seem to only be aware of above average levels of happiness in songs regardless of language

Native versus non-native

The last four plots on the right show the perceived valences of native speakers compared to those of non-native speakers. It seems that in general:

  • With the exception of Hindi songs, native speakers and non-native speakers did not differ much in their perceptions of valence in songs

  • Native Hindi speakers were less accurate than non-native speakers in perceiving the valence in Hindi songs according to Spotify API

  • Participants were better able to perceive the valence in Dutch and Mandarin songs (according to SPotify API) than Hindi and English songs, regardless of whether the songs were performed in their native language or not

  • Perceived valence was the most similarly rated between native and non-native speakers of English and Dutch

Column

Average Perceived Valence

Dutch Valence

English Valence

Hindi Valence

Mandarin Valence

Enjoyment and Familiarity

Column

Comparing enjoyment to familiarity

The graph on the right shows that there is a slight correlation between the familiarity and the enjoyment of a song; the more familiar, the more enjoyable. The most enjoyed songs were also the most familiar songs, namely the low valence English song, high valence Mandarin song, and high valence English song. Furthermore, it appears that the Dutch songs were the least familiar among our participants. All Dutch and Hindi songs were found to be the least enjoyable, regardless of the level of valence. It is remarkable that the Mandarin songs are almost as familiar and enjoyable as the English songs. Less surprising is the fact that the English songs were the most familiar and the most enjoyed, since all participants could speak English to at least some degree considering that the survey was written in English. However, it is interesting to see that the English song with the lowest valence was the most enjoyed, which could be explained by its high familiarity, since the perceived valence was low (see valence results).

Column

Familiarity and Enjoyment

Conclusion

Column

Discussion and Conclusion

The main goal of this study was to investigate the effect of native language on the perception of emotion in music. Additionally, the roles of familiarity and enjoyment in the perception of emotion in music were explored. The findings of the study revealed three key insights. First, our results indicated that being a native speaker of a language does not affect how a listener perceives emotions in songs. In fact, native and non-native speakers seem to mostly agree on the valence of songs. This finding concurs with previous research (Lee et al., 2021) that found that emotions in pop songs may be universally recognizable to listeners regardless of cultural background. Thus, it is likely that emotions in pop songs are universally recognizable to listeners regardless of native language.

Second, the level of valence has no effect on listener enjoyment. Although it may seem counterintuitive, happy songs are not necessarily more enjoyable. Instead, listeners could enjoy listening to sad songs (with low valence) more. This finding could be related to the concept of hedonism and eudaimonism. While the hedonistic view equates happiness with pleasure, the eudaimonic view equates happiness with feelings of meaningfulness obtained from solving complex human problems. For example, a sad song about grieving could be enjoyable because it helps its listeners make sense of their own grieving experience. This could help explain why listeners enjoy listening to sad songs as much as happy songs, sometimes even more.

Third, the more familiar a song is, the more enjoyable it is regardless of language and valence, and vice versa. This could be explained by the mere-exposure effect, which proposes that people develop a preference for something because they are more familiar with it. Furthermore, it is likely that one will intend to consume a piece of enjoyable music more frequently, and consequently develop a familiarity with it.

Fourth, it appears that the Spotify API algorithm is not accurately predicting valence values of Hindi songs, compared to songs in the other languages studied in this study. Native Hindi speakers in our study rated their perceived valence remarkably different from the algorithm, suggesting that Spotify API may be less sophisticated in capturing human perception of Hindi songs.

In conclusion, the present study provides some scientific and practical implications. While research in language and emotions in music is abundant, few have explored the role of native language in music perception. The findings of this study builds upon previous research and attempts to bridge this knowledge gap. Additionally, the findings of this study could help inform decisions in music marketing. For example, the genre “world music” that was created due to concerns that foreign music may be difficult to market because of language barriers may be irrelevant. Although a listener may not understand the lyrics of a song performed in a non-native language, our results show that they are capable of perceiving the emotions and enjoying the song as much as native speakers of the language. Therefore, based on the results of this study, it is recommended that music marketers rethink strategies for marketing songs in a foreign language.

Column

Limitations and suggestions for future research

Of course, this study is limited in some ways. First, the sample of songs consisted of songs from different decades (1985-2020), meaning that some songs were more recent than others. As a result, we cannot be sure if the variance in the dependent variables was solely caused by the presence or absence of a native language or by differences in song arrangement (or other song characteristics) over the years. Thus, it is recommended that future research control for song characteristics to reach more conclusive results.

Second, the sample of participants who are native Hindi speakers may not be large enough, making the related results less valid compared to those related to the other groups of participants. This was due to difficulties in recruiting native Hindi speakers, as the languages spoken in India are diverse and most Indians identify English as their native language. Future research could approach this problem by hiring research assitants in the local area.

Based on the findings of this study, a suggestion for future research would be to replicate this study using songs of a different genre. As it is suggested that pop songs may have intrinsically similar characteristics that are universally recognizable, comparing songs of a different genre (e.g. Rock, Rap, etc.) in different languages may yield interesting results that support or contradict the results of this study. For instance, it could be that genre characteristics affect perceived valence more than song lyrics.

Lastly, future research could attempt to investigate the role of native language on the perception of more complex emotions in music, such as grief and regret. Unlike basic emotions (e.g. happiness, sadness, etc.), complex emotions can vary across cultures in terms of appearance and require more cognitive processing (Brogaard, 2018). It is possible that the lyrics in songs can be cues to help listeners with the cognitive processing of complex emotions, such that native speakers of the language would be able to more easily perceive the complex emotions in the song than non-native speakers. Since complex emotions may manifest themselves differently across cultures, this could also be reflected in the lyrics of songs in different languages. Therefore, native language will potentially have an effect on the perception of complex emotions in music.